From ac4f70f6ce55c26e34fc80c5fabaa19cab077765 Mon Sep 17 00:00:00 2001 From: Chris Maffeo <cmaffeo2@illinois.edu> Date: Tue, 29 Nov 2016 11:05:43 -0600 Subject: [PATCH] Updated README --- README | 39 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 39 insertions(+) diff --git a/README b/README index 8706909..f002003 100644 --- a/README +++ b/README @@ -49,6 +49,10 @@ mkdir output You may use the '-g n' option to specify the n-th GPU on your machine, counting from 0. +If you fail to compile and link the applications, we recommend running +`make clean` to remove object files. Sometimes we have encountered +CUDA related errors in binaries built in a "dirty" environement. + /========\ | Citing | @@ -85,3 +89,38 @@ Max Belkin Emmanual Guzman Justin Dufresne Terrance Howard + + +/====================\ +| Outstanding issues | +\====================/ + +-- Features disabled during port to GPUs -- + +* Energy calculations are currently disabled + +* Constant chemical potential simulations + +-- Not implemented -- + +* Periodic boundaries are not handled by rigid body objects, so + simulations using these should only be performed for effectively + infinite systems + +* There are no checks to ensure that pairlists are recalculated before + particles further than the pairlist distance move to within the + cutoff + +-- Bugs -- + +* An issue with memory management causes some invalid memory to be + freed when the simulations end, cuasing a seg-fault; the impact of + this bug is mitigated because output has already been written at + this time + +* A large amount of GPU memory for pairlists is allocated statically, + which may cause out-of-memory crashes in older hardware + +* If the number of pairs in the sytem exceeds the length of the array + allocated for pairlists, the non-bonded kernel will try to access + forbidden regions of memory, causing a crash -- GitLab